All articulate assistants amalgamated, addressing assorted aspirations.
Assistance is an application to use and carry out conversations with multiple large language models simultaneously. It supports Bing Chat, Google Bard, ChatGPT3.5, and ChatGPT4.0. Assistance provides a way to get the most useful and accurate results from models that have been training on different datasets and for somewhat different tasks.
In the last few months, many different types of large language models (LLMs) have been released, including two versions of ChatGPT, Meta LLaMA, Google Bard, and Bing Chat. As each model was trained on different datasets, each one has its strengths and weaknesses. Instead of having to manually go through the major AIs, Assistance provides an intuitive way to query the models for information and switch between models while maintaining context.
- Be able to ask all models questions
- Allow users to select specific models they want to be able to continue talking to for more information
- Have the AIs remember context from previous questions you ask them
Gatsby Frameworkfor frontendMaterial UI, for stylingFastAPIfor backend- Reverse engineered
python APIsfor interacting with the LLMs
- First time using
GatsbyandFastAPI - APIs were reverse engineering and not offical, so documentation was very poor or nonexistent.
Through the development of Assistance, we learned a new frontend and server framework, along with how to use patchy or incomplete APIs.
Learned many small details regarding frontend and backend, such as:
- Converting strings to displaying markdown
- Advanced usage of contexes in React
- Importance of Memoization to prevent unnecessary rerenders
- Various ways to handle asynchronous code and manipulate data
- How to write custom docker containers
- Security of backend and frontend
- Firebase authentication for user accounts and multi device access through Firebase
- Ability for user to use their own API keys
- Paid subscription for more features